unclear: prepositions after nouns
A neuronOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in layerOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. 1 of Qwen3-0.6B. Its strongest snippets are mixed: they peak on different tokens. It is featuredA neuron picked for the full treatment: steering outputs and a written explanation. About 3,000 of the 86,000 neurons are featured., so it has steering outputs.
What this neuron does
This neuron weakly responds to small words like 'of', 'in', and 'to' that come right after a noun. The signal is faint and the steering results do not show a clear theme.
Written by Claude from this neuron's activations and steering outputs. Confidence: low. Think it is wrong? Flag it (opens a prefilled issue on GitHub).
Fires most on
The text windows where this neuron's activationThe number a neuron produces for a token. Large positive means it fired hard; negative means it pushed the other way. was highest. Amber marks positive activation; the outlined tokenA piece of text the model reads at once: often a word, sometimes part of a word or a punctuation mark. is the peak. Hover a token for its strength.
-
َّصَوُّف; صُوفِيّ ṣūfiyy / ṣūfī, مُتَصَوِّف mutaṣawwif), variously defined
-
mice into eight separate lines – four lines bred for high levels of daily running, with the
-
a living animal that brushed past the resin lost the hair, or the hair was lost in a similar manner by a mammal that came to feed on insects trapped in the resin, which later fossilised into amber.⏎The details of the discovery are
-
salmon before settling onto a suitable substrate are now extinct across most of their former range. Highland
-
frey, E.Godfrey@stfc.ac.uk⏎Research date: December 2008⏎E Godfrey et al., in Metal 07, proceedings of ICOM-CC metals group conference, Amsterdam (2
-
researchers, published in the Journal of Applied Ecology, asserts that poor standardisation of acoustic monitoring limits scientists’ ability to collate data.⏎Kate Jones, chairwoman of the UK-based Bat Conservation Trust⏎told the BBC that “without using the
SteeringAdding a push along a neuron while the model writes, to see how its output changes. Turning it up or down.
Push this neuron up or down while the model writes, and watch what changes. The outputs were generated ahead of time, so nothing runs in your browser.
Here is something I have been thinking about lately: "I am a student who is writing a research paper. I need help with some research paper writing help. I need help with some research paper writing help. I need help with some research paper writing help. I need help with some research
Most negative on
Where the activation went furthest below zero. Blue marks negative activation. Many neurons mean something different in each direction.
-
with actual tips and suggestions here’s some information to SSL certificates from their early days.⏎It was March 1995 when
-
is defined as all activity related to a player’s individual skill development. Based on these definitions, the following is a summary of USA Basketball’s stance on practice/training-to-competition throughout the four levels:⏎
-
path which you want to run using Run command. after selecting the path click on the next button.⏎Step 3: Set Run command⏎Now This is important to set run command.⏎Here you have to type the command name which you want to
-
worth considering at this time because, sooner or later, similar fights will break out in
Connections
Show on mapWhich neuronsOne of thousands of tiny dials inside each layer. It reads the text so far and produces a number, its activation, that nudges what the model writes next. in earlier layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. feed this one, and which later ones it feeds. Amber links excite, blue links suppress; longer bars are stronger.
Fed by earlier layers
- L0:1224 strongly suppresses
- L0:178 strongly suppresses
- L0:1301 strongly suppresses
- L0:1 suppresses
- L0:581 weakly excites
- L0:2791 weakly suppresses
- L0:386 weakly suppresses
- L0:83 weakly excites
- L0:214 weakly excites
- L0:1457 weakly excites
- L0:661 weakly suppresses
- L0:2195 weakly suppresses
- L0:95 weakly excites
- L0:29 weakly suppresses
- L0:639 weakly excites
- L0:621 weakly excites
- L0:799 weakly suppresses
- L0:915 weakly excites
- L0:1992 weakly suppresses
- L0:1174 weakly excites
- L0:1064 weakly suppresses
- L0:615 weakly suppresses
- L0:2840 weakly excites
- L0:531 weakly excites
Feeds later layers
- L15:888 excites
- L15:2616 excites
- L2:415 weakly suppresses
- L2:533 weakly suppresses
- L3:522 weakly suppresses
- L9:792 weakly excites
- L2:1307 weakly suppresses
- L9:1943 weakly excites
- L11:210 weakly excites
- L2:1139 weakly excites
- L2:1750 weakly suppresses
- L8:985 weakly suppresses
- L3:1173 weakly suppresses
- L22:1295 weakly excites 'as' before 'well' or 'as much'
- L2:1218 weakly suppresses
- L16:652 weakly excites
- L5:2825 weakly excites
- L6:1340 weakly suppresses
- L2:1206 weakly suppresses
- L3:2511 weakly suppresses
- L27:316 weakly suppresses
- L17:2413 weakly excites
- L5:2758 weakly suppresses
- L13:1411 weakly suppresses
Activation histogramA chart of how often a neuron takes each activation value.
How often this neuron fires at each strength. Almost always it does close to nothing (the tall bars in the middle); the snippets above are the rare moments it fired hard, amber for positive and blue for negative.
Logit lensReading which output tokens a neuron pushes toward or away from, straight off its output weights, before the later layers have their say.
The words this neuron nudges the model toward (amber) or away from (blue) when it fires. A hint, not a label: many neurons work through later layersOne step in the stack the text passes through. Qwen3-0.6B has 28 layers, each with 3,072 neurons. instead.
Promotes
- venience
- ница
- ;element
- /format
- ниц
- 啬
- �
- ETA
- iesz
- /+
Suppresses
- ited
- lành
- vũ
- uffer
- apol
- idences
- pups
- erge
- liğinde
- asymmetric